22

The Washington Post gives a detailed report of a recent study:

A stark new finding epitomizes that reality: In recent decades, rich black kids have been more likely to go to prison than poor white kids.

"Race trumps class, at least when it comes to incarceration," said Darrick Hamilton of the New School, one of the researchers who produced the study.

Does the study reasonably reach this conclusion?

  • 6
    Unless I'm mis-reading the article, the study didn't look at incarceration normalized by crimes committed/indiced for. Which, even if the study is accurate, makes its results remarkably meaningless. Less problematically, it is unclear if they normalized on other important factors like being in a single parent household (which has strong prediction to both crime AND income, and is/was disproportionately affecting non-whites, especially during time period noted). – user5341 Apr 6 '16 at 18:53
  • There seem to be other methodological issues (e.g. "Participants who were briefly locked up between interviews might not be included in their calculations of the share who were eventually incarcerated" would undercount people with shorter sentences). – user5341 Apr 6 '16 at 18:54
  • 3
    The article is available and easy to read given that their methodology wasn't very technical at all. They didn't appear to do anything at all about confounding factors - they didn't even wave their hands at them in the discussion section as far as I could see. – KAI Apr 6 '16 at 23:53
  • @user5341 if you are investigating whether or not one group of people is more likely to be incarcerated than another group of people, you should not normalize based on crimes the group of people are accused for. That may be on the causal pathway (one group may be more likely to be accused of a crime than another, and therefore more likely to be incarcerated). If, however, you are interested in conviction rate once accused (perhaps defining accused as charged with a crime), THEN you would use that as your population. It's a different question and a different claim. – De Novo Aug 24 '18 at 17:53
  • @user5341 the same thing can be said by type of crime. Lets say group A is more often charged with misdemeanors and group B is more often charged with felonies. If you're investigating the likelihood that members of group A vs group B will be incarcerated, you should not adjust for misdemeanors vs. felonies, because that may be on the causal pathway. It may be an important mechanism for group B being incarcerated more than group A -- they are more likely to be charged with a felony than members of group A. – De Novo Aug 24 '18 at 17:57
13

An examination of the methods shows, at best, the claim should probably be worded a little more carefully.

"In recent decades, rich black kids have been more likely to go to prison than poor white kids."

The measure of wealth

The measure of wealth in the study was total assets minus total debt for individuals in the cohort in 1985. At that time, the individuals in the cohort were 20-28 years old. This measure did include assets co-owned with a spouse, but, importantly did not include parental assets. "Rich kids" vs. "poor kids" probably doesn't accurately communicate the variable of interest here. For example, when I hear "rich kid", I'm more likely to think of parental wealth than individual wealth. Additionally, I would expect assets minus debt at 20-28 to be inversely correlated with education in many cases. Not going to college and working a full time low wage job, for example, would often give you more wealth during this age range than going to college, and education is associated with incarceration rate.

The measure of incarceration

Incarceration was measured indirectly. Interviews were conducted annually from 1979-1994 and biennially afterwards. The primary method for determining incarceration history was whether the respondent was incarcerated during the interview or interview attempt. There were direct questions in the 1980 survey, but responses to these questions would have only contributed to identifying prior incarceration (which disqualified a respondent from this analysis). Recall, this analysis looked at the relationship between individual wealth in 1985 and future incarceration.

Sample vs. population

The claim in both the Washington Post report of the article is ostensibly about the overall US population. The NLSY79 cohort is stated to be a nationally representative cross section. This is not supported with any data or a reference. This particular analysis uses a subgroup -- individuals who were not previously incarcerated. That seems like a reasonable choice, given the association between prior incarceration and both future incarceration and future wealth, and the goal of investigating the relationship between pre-incarceration wealth with future incarceration. However, because prior incarceration at 20-28 is also strongly associated with race, it is a choice that will bias the sample and prevent generalization to the overall US population.

Additionally, and probably most importantly, no inferential statistics were reported. The only measures reported were descriptive statistics of the sample. There is no confidence interval

A reasonable conclusion:

The only scientifically valid conclusion one can draw from this particular study is one about the sample itself. To make a claim about the overall US population using a representative sample, inferential statistics should be used. Any sample statistic comparing incarceration rates of groups based on race and wealth should include a confidence interval in order to draw conclusions about the population. If inferential statistics were applied to these data, the conclusion should be carefully worded to reflect the actual variables being measured, rather than simply rich, poor, and go to prison (vs. be in prison at any of several follow up interviews).

Importantly, though, despite not providing sufficient evidence for a reasonable conclusion, these data do suggest a possible racial component to incarceration that acts beyond an individual's wealth in young adulthood. It means the question should be evaluated more robustly, or, perhaps, that the analysis isn't finished. It may not be worthwhile to complete the analysis with these exact data, given the problems with how incarceration was defined.

A note on normalizing incarceration data by "crimes committed"

A highly upvoted comment suggested that the study should have "look(ed) at incarceration normalized by crimes committed". The truth of who has committed a crime and who hasn't, is, perhaps unfortunately, not an available statistic. It is generally not good practice to normalize for something you cannot accurately and precisely measure. It would be very nice if we could just know who committed a crime and who didn't, but in that world the criminal justice system would be very different indeed, and there would not be much need for this study or this question at all.

  • 2
    Welcome to the site, and congratulations on an excellent first answer! I do have one area I think could be improved: the last section on "normalizing by crimes committed" does not seem to explicitly connect to anything. I think it was likely in response to a comment on the question itself, but that connection is currently difficult to see and comments are frequently deleted. I would recommend starting that section with something like "a comment raised the idea of normalizing by crimes committed, but..." in order to clarify why it is being discussed. – Kamil Drakari Aug 27 '18 at 16:55
  • @KamilDrakari Thanks! I'll make the suggested edit when I get back from holiday and can sit down at my computer – De Novo Aug 31 '18 at 12:33
  • I think that “crimes committed” was meant to be “convictions by offenses”. For example are more rich black kids being convicted of drunk driving than poor white kids? It’s much easier to get convicted of drunk driving when you or your family own a car. Conversely, are more rich black kids being convicted of shop lifting than poor white kids? Social activist arrested while protesting or murder due to an argument over a remote? Whether or not they were guilty or not is a separate issue. Without knowing what they are convicted for, you can’t say what it means. Some kind of context is needed. – jmoreno Sep 15 '18 at 23:24
  • @jmoreno I'd refer you to my two comments on the question above. It is very bad statistical practice to normalize or adjust a statistic for an event that is on the causal pathway to the variable of interest. You would be asking an entirely different question by doing that. – De Novo Sep 16 '18 at 18:32

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .